文章摘要
基于机器学习法的儿童单侧腹股沟疝术后发生对侧腹股沟疝的预测模型构建
Prediction model construction of contralateral inguinal hernia after unilateral inguinal hernia operation in children based on machine learning method
投稿时间：2025-04-08  
DOI：10.3969/j.issn.1000-0399.2026.02.012
中文关键词: 儿童  单侧腹股沟疝术  对侧腹股沟疝  预测模型  机器学习
英文关键词: Children  Unilateral inguinal hernia  Contralateral inguinal hernia  Prediction model  Machine learning
基金项目:
作者单位
姚丹丹 473000 河南南阳 南阳市中心医院小儿外科 
赵成鹏 473000 河南南阳 南阳市中心医院小儿外科 
段永福 473000 河南南阳 南阳市中心医院小儿外科 
刘梦磊 473000 河南南阳 南阳市中心医院磁共振影像科 
摘要点击次数: 290
全文下载次数: 58
中文摘要:
      目的 基于机器学习方法构建儿童单侧腹股沟疝术后对侧腹股沟疝（MCIH）发生风险的预测模型，并对预测效能进行验证。方法 收集2021年1月至2023年4月南阳市中心医院252例行单侧腹股沟疝术的儿童临床资料，作为建模组，根据术后是否发生MCIH分为发生组（n=36）和非发生组（n=216），按照建模组与验证组7∶3的比例于2023年5月至2024年6月另选取患儿108例作为验证组。使用logistic回归、决策分类回归树（CART）、反向传播神经网络（BPNN）的机器学习算法构建儿童单侧腹股沟疝术后MCIH发生的预测模型，并采用受试者工作特征（ROC）曲线比较3种方法构建的模型对术后发生MCIH的预测价值。结果 单因素及多因素结果显示，年龄、性别、低体质量儿、早产儿、家族性腹股沟疝病史、疝囊位置、术后并发症是患儿术后发生MCIH的独立影响因素（均P<0.05）。构建的logistic回归、CART、BPNN模型的准确度分别为83.76%、72.61%、81.50%，灵敏度分别为88.90%、88.90%、83.30%，特异度分别为82.90%、69.90%、81.20%。ROC曲线分析显示，logistic回归、CART、BPNN模型的曲线下面积（AUC）分别为0.918、0.862、0.899（均P<0.05）。3种机器学习算法构建的模型AUC均>0.800，预测准确性良好，其中logistic回归模型的AUC及准确度均高于CART和BPNN模型，其预测患儿单侧腹股沟疝术后发生MCIH的效能最优。结论 儿童单侧腹股沟疝术后发生MCIH的影响因素有年龄、性别、低体质量儿、家族性腹股沟疝病史等，基于机器学习算法构建的儿童单侧腹股沟疝术后发生MCIH的预测模型均具有较好的预测效能，其中以logistic回归模型预测效能最佳，具有较高的预测准确性。
英文摘要:
      Objective To construct the risk prediction model of contralateral inguinal hernia(MCIH) in children after unilateral inguinal hernia surgery based on machine learning method, and to verify the prediction efficiency. Methods The clinical data of 252 children undergoing unilateral inguinal hernia operation in Nanyang Central Hospital from January 2021 to April 2023 were collected. As a modeling group, patients were divided into the occurrence group(n=36) and non-occurrence group(n=216) according to whether MCIH occurred after operation. According to the ratio of 7:3 between the modeling group and the verification group, 108 additional children were selected as the verification group from May 2023 to June 2024. Machine learning algorithms of logistic regression, decision classification regression tree(CART) and backpropagation neural network(BPNN) were used to construct a prediction model for postoperative MCIH in children with unilateral inguinal hernia, and receiver operation(ROC) curve was used to compare the prediction value of the models constructed by the three methods for postoperative MCIH. Results Univariate and multivariate results showed that age, sex, low body mass infants, premature infants, family history of inguinal hernia, location of hernia sac and postoperative complications were independent influencing factors for postoperative MCIH(all P< 0.05). The accuracy of logistic regression, CART and BPNN models was 83.76%, 72.61% and 81.50%, respectively, the sensitivity was 88.90%, 88.90% and 83.30%, respectively, and the specificity was 82.90%, 69.90% and 81.20%, respectively. ROC curve analysis showed that the area under the curve(AUC) of logistic regression, CART and BPNN models was 0.918, 0.862 and 0.899, respectively(all P< 0.05). The AUC of the models constructed by the three machine learning algorithms were all> 0.800, and the prediction accuracy was good. The AUC and accuracy of logistic regression model were higher than those of CART and BPNN model, and it had the best effect in predicting MCIH after unilateral inguinal hernia in children. Conclusion The influencing factors of postoperative MCIH in children with unilateral inguinal hernia include age, gender, low body mass, family history of inguinal hernia, etc. The prediction model of postoperative MCIH in children with unilateral inguinal hernia based on machine learning algorithm has good prediction efficiency, and the logistic regression model has the best prediction efficiency with high prediction accuracy.
查看全文   查看/发表评论  下载PDF阅读器
关闭